2.4.19 rmk4 Jffs2 Performance problems, Upgrade diffulculty.

Hello,

I have been a content user of the kernel version 2.4.19 rmk4 for
several months. Everything has been working quite well.
My configuration is a Arm720t processor, 8 meg of 32 bit wide NOR
Flash, 16 meg of ram ( 4 meg ramdisk).

Half of the Flash is reserved for a JFFS2 file system.

The problem that I have encountered has been the time it takes to
write a single large file into the JFFS2 file system. It takes approximately 4 minutes to write a single 1 meg file into the file system. Writing a second file of the same size takes approximately 8 minutes. Each time I add another large file, it takes progressively longer. I am seeing about a 60% compression on those files.

I thought that this performance may have been improved with the
current version of the jffs2 files so I got the tar files, mtd-snapshot-20030305.tar, and updated my configuration. Now I am not able to copy any large files.
I am running an almost pristine copy of 2.4.19 rmk4.

I have applied the patch, linux-2.4.19-pre10-shared-zlib.

I have updated to mtd-snapshot-20030305.tar.

Using the latest mkfs.jffs2.
This is the sequence of events

# mount -t jffs2 /dev/mtdblock0 /home
# cp test_file home
Unable to handle kernel paging request at virtual address e5962018
pgd = c06e4000
*pgd = 00000000, *pmd = 00000000
Internal error: Oops: ffffffff
CPU: 0
pc : [<c0136ac8>]    lr : [<c0136794>]    Not tainted
sp : c06e9d88  ip : 00000000  fp : 00000000
r10: c07c7ba5  r9 : 00000009  r8 : 00000015
r7 : 00000005  r6 : c01749bc  r5 : c186401c  r4 : e5962008
r3 : c1865240  r2 : c1864a40  r1 : 00000005  r0 : 00000008
Flags: nZCv  IRQs on  FIQs on  Mode SVC_32  Segment user
Control: 217F  Table: C06E4015  DAC: 00000015
Process cp (pid: 23, stack limit = 0xc06e8368)
Stack: (0xc06e9d88 to 0xc06ea000)
Stack: (0xc06e9d88 to 0xc06ea000)
9d80:                   00000000 c0136794 c01749bc c06e9db0 c06e9dac 
c1864a40
9da0: c01749bc c1867740 000000a3 c1865240 c1864a40 00000005 00000008 
00000000
9dc0: 000acc48 c0015b5c 000000e8 c01749bc fffffffb 00000005 00000000 
c07c7b60
9de0: 00000000 00000000 c0d54b60 c0137a20 c07c7b60 000000e8 fffffff1 
c00a0428
9e00: 000000e8 c07c7a60 c06a1100 c009f778 c00a451c 00000100 c07c7b60 
00000100
9e20: c069d000 000000e8 c06d85f0 c0d54b60 00001000 00000000 c06efe7c 
c069d000
9e40: c07b04bc 00002000 c01a2318 c00a475c 00000000 c01a2318 c01a2330 
c06a4ab0
9e60: c07b04bc 00000000 00001000 c06a49a0 c00a2884 00001000 00000100 
00000000
9e80: c06a4ab0 c06a49a0 c00a2d6c c017c84c 0000075a c00c40e0 c06a49e4 
00001000
9ea0: 00000000 00000100 00000000 00000002 00000000 00000400 c00c2800 
c06e9f04
9ec0: 00000000 00000001 00000004 c0063890 000003ec c008568c c0d27b00 
20000013
9ee0: c0160228 00000041 00000000 c015fff4 000001d2 00000100 c0160224 
00000100
9f00: c0162c64 c01a2318 00001000 00000000 c069d000 00000100 c06a49a0 
00000100
9f20: c00553c0 c0053604 c06e9f5c 00000100 c01a2330 c001d0ec 00000000 
c06a4a08
9f40: fffffff4 00000000 c06a4a44 c07b25c0 bffffc2c c07b25a0 00000000 
1a000100
9f60: c07b25c0 c07b25a0 ffffffea 00000000 00000100 c06e8000 bffffc2c 
00000000
9f80: c0060638 c0170860 c00453dc 00000100 00000100 00080f80 00000004 
c0036724
9fa0: 00080000 c00365a0 00000100 c00363e8 00000004 bffffc2c 00000100 
0004a618
9fc0: 00000100 00000100 00080f80 bffffc2c 00000100 bffffc2c 00080000 
00000000
9fe0: 00006050 bffffbd4 0004a310 00053c84 20000010 00000004 00000000 
00000000
Backtrace: no frame pointer
Code: e92d4010 e59d4004 e5944020 e3a0c000 (e5c40010)
Segmentation fault
#
Piece of my Ssytem.map
    c0135f70 T zlib_inflate_blocks_reset
    c0135fe8 T zlib_inflate_blocks_new
    c0136034 T zlib_inflate_blocks
    c0136a68 T zlib_inflate_blocks_free
    c0136a7c T zlib_inflate_set_dictionary
    c0136aa4 T zlib_inflate_blocks_sync_point
    c0136ab8 T zlib_inflate_codes_new
    c0136ae4 T zlib_inflate_codes
    c013717c T zlib_inflate_codes_free

When I reboot I get this error.

# mount -t jffs2 /dev/mtdblock0 /home
jffs2_scan_eraseblock(): Magic bitmask 0x1985 not found at 0x000c0000: 
0x2003 id
jffs2_scan_eraseblock(): Magic bitmask 0x1985 not found at 0x000c0004: 
0x000c id
jffs2_scan_eraseblock(): Magic bitmask 0x1985 not found at 0x000c0008: 
0xdc6d id
jffs2_scan_eraseblock(): Magic bitmask 0x1985 not found at 0x000e0000: 
0x2003 id
....
# // at this time, I cannot write anything to the jffs2 filesystem.

The oops error seems to be consistently at zlib_inflate_codes_new + 0x10.

Have I missed something? Before I updated, I could very reliably write files, albeit very slow for the large file (>1 meg.). Now I am unable to write reliably at all.

Thanks,
RobertS

Could you please try the following:
1. Get your almost pristine copy of 2.4.19.rmk4
2. use latest MTD snapshot and use patchin.sh from patches directory. This will apply zlib patch too
3. Report results

Thomas
________________________________________________________________________
linutronix - competence in embedded & realtime linux
http://www.linutronix.de
mail: tglx at linutronix.de

Hello Thomas,

Sure.

Please provide the name and location of the MTD snap shot so I use
the correct one. The mtd snapshot that I used originally was mtd-snapshot-20030305.tar.gz. I had numerous problems with this snap shot and 2.4.19. Unfortunately, there was a lot of "pilot" error when I started this upgrade effort. Let me know and I will proceed carefully and capture the results.

The performance of approximately 4 minutes to copy a 1 meg file was
with an almost "pristine" 2.4.19 rmk4 kernel. Does this meet expectations for that iteration of jffs2?

Robert

cvs -d :pserver:anoncvs at cvs.infradead.org:/home/cvs login (password: anoncvs) cvs -d :pserver:anoncvs at cvs.infradead.org:/home/cvs co mtd

I'va never seen such long times

Thomas
________________________________________________________________________
linutronix - competence in embedded & realtime linux
http://www.linutronix.de
mail: tglx at linutronix.de

The update and build went smoothly. Unfortunately, I still oops in exactly the same manner.

What I am doing now is turning on the verbose debug messages and keep everything as simple as possible. I will be performing a series of tests and will use this thread either to post my findings or further questions.

RobertS.

Set JFFS2_DEBUG=1 and do an
echo 9 >/proc/sys/kernel/printk
before you mount the device and start copying.
Log via a serial console if possible, as klogd seems to loose bits, if there is heavy load.

Send this captured log to me in private or put it anywhere for download.

Thomas
________________________________________________________________________
linutronix - competence in embedded & realtime linux
http://www.linutronix.de
mail: tglx at linutronix.de